fallout of lead over paris from the 2019 notre-dame...

manuscript submitted to GeoHealth

Fallout of Lead over Paris from the 2019 Notre-Dame1

Cathedral Fire2

Alexander van Geen1, Yuling Yao2, Tyler Ellis1, and Andrew Gelman23

1Lamont-Doherty Earth Observatory of Columbia University. Palisades, NY 10964, USA.42Department of Statistics, Columbia University. New York, NY 10027, USA.5

Key Points:6

• Surface soil Pb concentrations within 1 km of Notre-Dame cathedral are about7

200 mg/kg higher downwind of the fire relative to background.8

• The corresponding fallout of 1000 kg Pb is 6 times higher than the estimated mass9

of Pb from the fire transported by the wind beyond 1 km.10

• The resulting human exposure was probably dwarfed by the impact of leaded-gasoline11

in previous decades but warranted more testing sooner.12

Corresponding author: Alexander van Geen, [email protected]

–1–


Abstract13

The roof and spire of Notre-Dame cathedral in Paris that caught fire and collapsed on14

April 15, 2019, were covered with 460 tons of lead (Pb). Government reports documented15

Pb deposition immediately downwind of the cathedral and a 20-fold increase in airborne16

Pb concentrations at a distance of 50 km in the aftermath. For this study, we collected17

100 samples of surface soil from tree pits, parks, and other locations in all directions within18

1 km of the cathedral. Concentrations of Pb measured by X-ray fluorescence range from19

30 to 9000 mg/kg across the area, with a higher proportion of elevated concentrations20

to the northwest of the cathedral, in the direction of the wind prevailing during the fire.21

By integrating these observations with a Gaussian process regression model, we estimate22

that the average concentration of Pb in surface soil downwind of the cathedral is 430 (95%23

interval, 300-590) mg/kg, nearly double the average Pb concentration in the other di-24

rections of 240 (95% interval, 170-320) mg/kg. The difference corresponds to an inte-25

grated excess Pb inventory within a 1 km radius of 1.0 (95% interval, 0.5-1.5) tons, about26

0.2% of all the Pb covering the roof and spire. This is over 6 times the estimated amount27

of Pb deposited downwind 1-50 km from the cathedral. To what extent the concentrated28

fallout within 1 km documented here temporarily exposed the downwind population to29

Pb is difficult to confirm independently because too few soil, dust, and blood samples30

were collected immediately after the fire.31

Plain Language Summary32

This study estimates the extent to which the population of Paris was exposed to33

lead as a result of the Notre-Dame cathedral fire of April 15, 2019. The concern stems34

from the large quantity of lead that covered the cathedral, some of which was injected35

into the air by the fire for several hours. In order to evaluate how much lead rising from36

the fire was redeposited nearby, surface soil samples were collected in all directions from37

tree pits and parks within a 1 km radius of the cathedral. Elevated levels of lead observed38

downwind of the cathedral indicate that surface soil preserved the mark of lead fallout39

from the fire. Although the estimated amount of lead redeposited within 1 km corresponds40

to only a small fraction of the total covering the cathedral, it could have posed a health41

hazard to children located downwind for a limited amount of time. Environmental test-42

ing on a larger scale immediately after the fire could have provided a more timely as-43

sessment of the scale of the problem and resulted in more pointed advice to the surround-44

ing population on how to limit exposure to the fallout of lead.45

1 Introduction46

The roof and spire of Notre-Dame cathedral in the center of Paris covered with 46047

tons of lead (Pb) tiles burned down within a few hours of a fire that started early on the48

evening of April 15, 2019, and took 9 hours for the fire brigade to extinguish (INERIS,49

2019). The yellow color of the smoke rising from the cathedral during the first few hours50

has been attributed to PbO particles entrained with the hot ascending air and formed51

by heating to 600◦C the lead on top of the vault of the cathedral. Prevailing winds com-52

bined with modeling of the plume of smoke particles rising from the fire have linked this53

increase to the ejection of about 150 kg of Pb, only 0.03% of the total covering the cathe-54

dral, into the atmosphere by the fire and redeposition over several tens of kilometers. This55

is consistent with observations at an air quality monitoring station 50 km downwind of56

the burning cathedral where a 20-fold increase in particulate Pb concentration, from 0.05057

to 0.105 µg/m3, was recorded during the week that followed the fire (Fig. 1a). The same58

INERIS (2019) report also states that considerably more Pb was likely deposited in the59

immediate vicinity of the cathedral but there was no attempt to estimate this amount.60

The sequence of announcements and measures taken after the fire by local author-61

ities provide a context for and contribute to the interpretation of the new Pb measure-62

–2–


ments presented here. The consequences of the Notre-Dame fire are worth document-63

ing because lead has neurotoxic effects even at low levels of exposure at a young age (Lan-64

phear et al., 2005; Laidlaw and Filippelli, 2008; Aizer and Currie, 2019). Dust and soil65

are also known sources of child Pb exposure, including in France (Etchevers et al., 2015;66

Glorennec et al., 2016). Four days after the fire, on April 19th, the environmental non-67

governmental organization Robin des Bois (2019) issued a press release expressing con-68

cern about the likely large quantities of Pb mobilized by the fire, referring to potential69

health risks incurred by firefighters, workers on the site, and the surrounding population.70

On April 27th, almost two weeks after the fire, the Agence Regionale de la Sante (ARS,71

2019a) co-issued a press release indicating that dust sampling had revealed some locally72

elevated levels of Pb and that areas very close to the cathedral that could not rapidly73

be cleaned had been closed to the public. The press release also recommended that nearby74

inhabitants remove indoor dust with wet wipes and announced follow-up studies to min-75

imize risks to workers on the site and the surrounding population. On May 9, 2019, the76

ARS (2019b) confirmed soil Pb levels of 10,000-20,000 mg/kg in the out-of-bounds area77

very near the cathedral but also reported that no levels above 300 mg/kg, the maximum78

level recommended in France (HCSP, 2014), were measured outside this area within the79

Ile de la Cite, where the cathedral is located. The same news release from the ARS re-80

ported that no sample collected around the cathedral to assess air quality exceeded the81

regulatory level of 0.25 µg/m3 for Pb in airborne particulate matter. This indicated that82

exposure through inhalation was unlikely, although the timing of the sampling relative83

to the fire was not provided.84

Figure 1: Events following the April 15, 2019 Notre-Dame cathedral fire shown with (a)weekly time series of Pb concentrations in airborne particulate matter measured at twoAirparif monitoring stations (https://www.airparif.asso.fr/en/) and (b) the total numberof children and adolescents in the 1st, 4th, 5th, and 6th arrondissements whose blood wastested for Pb (ARS, 2019g, h).

–3–


Almost a month later, on June 4th, the ARS (2019c) reported that indoor dust col-85

lected in some nearby apartments was found to be elevated in Pb and referred to a first86

child whose blood-Pb content was over 50 µg/L, the local intervention level requiring a87

follow-up investigation at home (HCSP, 2014). In the same press release, whose over-88

all tone was meant to be reassuring, the ARS offered to test the blood of any children89

less than 7 years old residing on the Ile de la Cite for Pb at a nearby hospital. On July90

18, 2019, the ARS (2019d) issued a 100+ page report indicating no blood-Pb levels above91

50 µg/L had been detected in 81 children from the 1st, 4th, 5th, and 6th arrondissements,92

all areas downwind of the fire, and that a Pb source unrelated to the fire was identified93

in the home of the previously reported child with >50 µg/L Pb in blood. The same doc-94

ument indicated that indoor surface Pb concentrations at a number of nurseries sam-95

pled downwind of the fire were all <1000 µg/m2 , the local regulatory level after lead96

remediation in housing, and mostly <70 µg/m2, the level above which a blood test is en-97

couraged (HCSP 2014), along with a detailed map of measurements of Pb concentrations98

in surface dust of the area. Unlike soil measurements, which require unconsolidated ma-99

terial such as a tree pit or a park, surface Pb measurements, usually conducted indoor,100

rely on wiping a hard surface (e.g. a sidewalk) over a set area with a wet tissue that is101

then analyzed. This is a standard regulatory procedure in France as well as in the U.S.102

(Lanphear et al., 1995; JORF, 2009).103

On July 26, Robin des Bois filed a lawsuit claiming insufficient measures were taken104

to protect the health of workers on the cathedral site, after which activities were inter-105

rupted for several weeks (Le Monde, 2019). Soon thereafter on August 4, the ARS (2019e)106

tried to refute allegations by Mediapart (2019), an investigative online news provider,107

that it was minimizing the risk of Pb exposure to the population residing downwind of108

the cathedral. On November 27, however, the ARS (2019f) announced online access to109

georeferenced environmental Pb data collected both before and after the cathedral fire110

(https://santegraphie.fr/mviewer/?config=app/notredame od.xml). The data posted111

by the ARS include a dozen wipe-based surface Pb measurements conducted in 2018 in112

close proximity to the cathedral and about 60 measurements of the same type in the same113

area from 2020. In the 2018 and 2020 data, only one measurement exceeds 5000 µg/m2114

Pb, and this by less than a factor of two.115

For 2019, the database contains a much larger number of measurements around the116

cathedral, including dozens extending over a distance of 50 km in the direction of the117

plume and the air-quality monitoring station of Limay where an increase in airborne Pb118

had been detected during the week after the fire (Fig. 1). Outside a radius of 2 km from119

the cathedral, none of the reported measurements exceed 5000 µg/m2. Between 1 and120

2 km from the cathedral, a subset of 7 out of a total of ∼40 measurements, all conducted121

between mid-May and mid-June 2019, exceed 5000 µg/m2 and but in all but one case122

by less than a factor of 10. Within a radius of 1 km of the cathedral, the proportion and123

level of elevated surface Pb measurements is comparable to the findings in the 1-2 km124

range, although the majority of these measurements date from summer and fall 2019,125

i.e. several months later. It is only within a radius of 100 m from the cathedral that much126

higher surface Pb concentrations, most over 100,000 µg/m2 and several near 1,000,000127

µg/m2 are reported on the ARS site.128

The ARS georeferenced data site only lists 24 soil Pb measurements within a ra-129

dius of 2 km from the cathedral, all conducted after the fire and between April and June130

2019. Most of the reported Pb concentrations are below 100 mg/kg, with 6 in the 100-131

300 mg/kg range, and only one higher value of 310 mg/kg within 100 m of the cathe-132

dral. These values do not seem consistent with the 10,000-20,000 mg/kg concentrations133

reported for the same area by the ARS (2019b), which were not posted, unless the mea-134

surements were obtained by different methods. The soil protocol followed by the ARS135

calls for sampling to 5 cm depth and homogenizing this material before analysis. In the136

case of Pb contamination limited to the top 1 mm, this could lead to >50-fold lower con-137

–4–


centrations than measured from the very surface with a hand-held XRF fluorescence an-138

alyzer (Landes et al, 2019). Diluting the highest reported surface Pb concentration of139

1,000,000 µg/m2 over the mass of soil to 5 cm depth would, for instance, increase the140

soil Pb concentration by only 10 mg/kg, i.e. little more than 10% of background levels141

based on the other measurements. The relatively low soil concentrations posted on the142

ARS site are therefore not necessarily inconsistent with the much higher levels referred143

to in the earlier press release.144

One of the goals of this study was to determine if a very basic soil sampling pro-145

cedure of the fallout paired with more advanced statistical analysis could yield useful in-146

formation. Provided sampling is limited to the top ∼1 cm, soil has the advantage of pre-147

serving the signal of a fallout for much longer than hard surfaces such as road and side-148

walks that are swept by wind and flushed by rain. Our surface soil data collected 9-10149

months after the fire show that the population residing within 1 km and downwind of150

the fire was probably considerably more exposed to Pb fallout, albeit for a brief period,151

than indicated by measurements and surveys conducted by local authorities weeks to months152

later. The study demonstrates that the public should expect data to be collected and153

offered to scrutiny immediately after an environmental accident. Besides knowing where154

a potential hazard is located, posting of data creates incentives for local authorities to155

act transparently and in the public interest. Other cases, albeit of a very different mag-156

nitude, where lack of data unnecessarily diminished public trust and may have led to the157

wrong official responses include the nuclear reactor accidents in Chernobyl and Fukushima158

(Alexievich, S., 2006; Brown et al., 2016).159

2 Materials and Methods160

2.1 Data collection161

One hundred soil samples were collected between December 20, 2019 and Febru-162

ary 29, 2020 mostly from tree pits (55 samples) and parks or smaller garden-like areas163

(30). In a few cases, samples were collected from small gaps in the sidewalk (13) or even164

semi-permanent plant pots (2) for lack of more suitable alternatives. One set of 58 soil165

samples were spaced roughly equally along two concentric circles of 400 and 1000 m in166

radius centered on the cathedral (Fig. 2). The remaining 42 samples targeted the area167

likely to have been impacted by fallout from fire, downwind of the cathedral.168

A large metal spoon was used to recover ∼50 g of material from the upper ∼1 cm169

of each site. The samples were dried overnight in paper bags, after which the fine frac-170

tion was separated through a metal kitchen sieve (∼1 mm mesh size) and poured into171

20 mL scintillation vials. Without further processing, the fine fraction was analyzed in172

the inverted vials through plastic cling wrap using a handheld Innov-X (now Olympus)173

Delta Premium X-ray fluorescence analyzer. The XRF’s internal calibration was con-174

firmed by bookending both rounds of analyses with Standard Reference Material soil 2711a175

from the US National Institute of Standards and Technology. The average of 1, 480±176

40 mg/kg (n = 4) obtained for Pb was consistent with the certified value of 1400±10177

mg/kg.178

The XRF measures the concentrations of 16 additional elements. Tin (Sn) is of par-179

ticular interest for the present study but there is no certified Sn value for SRM 2711a.180

Landes (2019) compared soil Sn concentrations measured by the same instrument with181

two dozen soil digests analyzed by inductively-coupled plasma mass spectrometry (Cheng182

et al., 2004). The slope of Sn concentrations measured by XRF as a function of concen-183

trations measured by ICPMS of 1.64 indicates a systematic overestimate of Sn concen-184

trations by XRF.185

Soil Pb concentrations are also displayed in a polar coordinate centered on the cathe-186

dral to help to visualize the impact of the fire independently of the presumed direction187

–5–


48.84

48.85

48.86

2.33 2.34 2.35 2.36

longitude

latit

ude

Soil Pb (ppm) <100 100−300 300−1200 >1200

Figure 2: Map of 100 soil sample locationsaround the cathedral and their Pb concen-trations. The two circles of samples centeredon the cathedral have radius of 400 and 1000m, respectively. Additional samples were col-lected in downwind direction, northwest of thecathedral.

0.0

0.2

0.4

0.6

<100 200 300 600 900 >900

Pro

port

ion

outside

inside plumes

n=20

n=10

n=5

Soil Pb (ppm)

0.0

0.2

0.4

0.6

<100 200 300 600 900 >900

Pro

port

ion crack

garden

parkpot (all<100)

tree

Figure 3: (a) Proportion of soil Pb col-lected inside and outside the area passedover by the plume of smoke rising fromthe cathedral. (b) Proportion of Pb sam-ple for different types of soils. The sizeof the symbols indicates the number ofsamples in each grouping. The two plantpots are low in Pb and their symbol outof range.

●●● ●

●

●●

●

●

●

●

●

●●● ●

●

●●●

●

●

●

●●

●

●

●

●● ●

●

●●

●●

●

●●

●

●

●

●

●

●

●

●

●

●

●

●

●

●●

●

●●●●

●

●

●

●

●

●

●

●●

●

●

●

●

●

●

●

●

●

●

●

●

●●

●●

●

●

●

●

●●

●

●●

●

●

●●

●

●

●

●ND

Sampled values

x = inside plume

Polar coordinates

0 1000 2000N

E

S

W

N

distance (m)

bearing from the

cathedral

101

102

103

104

●

●●

●

●● ●●

●●

● ●● ● ●● ● ● ●●● ●●●●●● ●●

●● ●● ●● ● ●●● ● ●●● ● ●●● ●● ●● ●● ●● ●●●●● ● ●●● ● ●● ●● ●●● ●●●

●●● ● ●● ●● ● ●●●● ●● ●● ●● ●● ●●●●

●

●

●

●

●

crackgardenparkplant pottree

0 1000 2000

10

100

1000

10000

Soil Pb by distance to the center

distance (m)

Soil Pb

ppm

●

●●

●

●● ●●

●●

●● ●●● ●●● ●●●●● ●●● ●● ●

●●● ●●● ● ●● ●●● ●● ● ●● ●● ●● ● ●●●●●● ●● ● ●● ●● ●● ●● ●● ● ●●● ●

●● ●●● ●● ●●● ● ●● ●●●● ●● ●●●● ●●

0(N) 90(E) 180 270 360

10

100

1000

10000Soil Pb by angle to the center

angle (North=0)

Soil Pb

ppm

0 1000 2000

10

100

1000

10000

Soil Pb by distance to the center, inside−plume samples

distance (m)

● original circlesecond sample

0 1000 2000

10

100

1000

10000Soil Pb by distance to the center,

out−of−plume samples

distance (m)

Figure 4: Left column: Sampled locations and Pb concentrations in both Cartesian andpolar coordinates. Middle column: scatter plot of soil Pb by distance and bearing fromthe cathedral, colored by soil type. Right column: soil Pb by distance from the cathedral,grouped by inside/outside plume.

–6–


of the plume (Fig. 4). The sampled Pb peaks at the northwest, and as a whole, drops188

off with a longer radial distance, while the slope inside the plume is sharper. Based on189

INERIS (2019), we specify the plume region to be the sector between 260° to 310° clock-190

wise from the cathedral (Fig. 4).191

2.2 Notation and pre-processing192

We denote the soil Pb concentration (in mg/kg) in the i-th location to be yi, i =193

1, . . . , n, and compute the its radical distance ri (in km) and the bearing θi (in degrees,194

North = 0, East = 90) from the cathedral. We index the type of soil by k[i] ∈ {1, 2, . . . , 5}195

to represent where the i-th sample was drawn from: cracks in the sidewalk, smaller gar-196

den areas, park, plant pots, or tree pits.197

As for many other natural measurement, the observed yi has a heavy right tail. Di-198

rectly modeling y will often make the model sensitive to the few extreme values. Mea-199

surement errors in chemical analysis are often additive in the lower end and multiplica-200

tive in the high end. Therefore instead of a log transformation, we prefer a 1/4-th power201

transformation, as the measurement errors would likely be of similar order of magnitude202

in different sites. For notation simplicity, we will plug in y1/4 → y in the model descrip-203

tion we use, and transform it back to the ordinary scale after model fitting.204

The concentration of soil Pb varies both spatially and by soil type. We decomposethe outcome yi into three terms:

yi = µk[i] + f(ri, θi) + εi, εi ∼ N(0, σ2), i = 1, ..., n.

which includes205

• the soil type coefficient µk[i]; it depends only on what type of soil the sample be-206

longs to;207

• the spatial term f(ri, θi); it depends only on where the sample is collected (en-208

coded by distance and bearing);209

• an independent observational noise εi; it contains measurement error, small-scale210

fluctuations, and effects from any unmeasured covariates.211

2.3 Hierarchical modeling of different soil types212

The lower row of Fig. 3 and the middle column of Fig. 4 suggest that the type of213

soil (tree pit, park, smaller garden areas, cracks in the sidewalk, plant pots) is predic-214

tive of Pb concentrations. The sample sizes in different types are unbalanced, and a sim-215

ple sample mean is noisy for groups with small samples. To partially pool across the data,216

we fit a hierarchical model to the soil type coefficients µk.217

However the model is not identifiable yet, as a additive constant can be extracted218

from the µ and added to f . To resolve this, we restrict the soil-type coefficients by a zero-219

sum constraint,∑5k=1 µk = 0.220

2.4 Modeling the Pb distribution by a Gaussian process regression221

We model the spatial pattern nonparametrically by placing a mean-zero Gaussianprocess prior on the latent function f . It models the joint distribution at any two loca-tions, f(r, θ), f(r′, θ′), using a multivariate Gaussian distribution. To flexibly accountfor the influence from the distance, bearing, and their interactions, we use a product ker-nel in the Gaussian process prior:

K(r1, θ1, r2, θ2) := Cov(f(r1, θ1), f(r2, θ2)) = αK1(r1, r2)K2(θ1, θ2),

–7–


where for distances, we adopt the commonly-used squared exponential kernel:

K1(r1, r2) = exp

(− (r1 − r2)2

ρ2r

).

For the bearing, we employ a periodic kernel:

K2(θ1, θ2) = exp

(−2 sin2(π|θ1 − θ2|/360)

ρ2θ

).

Besides the soil type effect µ, spatial latent function f , and scale of the observa-222

tional variation σ, the model also contains hyperparameters α: the scale of the spatial223

signal (how strong the spatial pattern is); ρd: the length scale in the distance dimension224

(how rigid the function f can change over distance); and ρθ: the length scale in the an-225

gle dimension.226

Since the modeled outcome y1/4 and the distance (in km) are all roughly unit-scaled,we adopt weakly informative priors

ρd ∼ N(0, 1.52), ρθ ∼ N(0, 1), α, σ ∼ N(0, 62), µk ∼ N(0, 1), k = 1, . . . , 5.

We sample from the posterior distribution of all parameters in the model using Stan227

(Stan Development Team, 2018). In our example, the chains mixed well for 4 chains and228

3000 iterations per chain.229

2.5 Inference from the fitted model230

2.5.1 Spatial imputation from the model231

We sample a uniform 30×30 grid of locations (ρ, θ) in the 1.5 km neighborhood.After integrating out the posterior distribution, we obtain the posterior predictive dis-tribution of the outcome values at this location f = f(ρ, θ) is from

f |ρ, θ, ρ, θ, f ∼ N(K(ρ, θ, ρ, θ)K−1(ρ, θ)f,K(ρ, θ)−K(ρ, θ)K−1(ρ, θ)K(ρ, θ, ρ, θ)

). (1)

We model the outcome to the 1/4 power and transform f back to f4 in the visualiza-232

tions.233

Further, we add the observational noise and generate the posterior predictive dis-tribution of y outcome y in location (ρ, θ) by location f = f(ρ, θ) is from

y|f ∼ N(f |σ2sim), σsim ∼ p(σ|y). (2)

This amount to the prediction of the outcome in a typical soil type with µ = 0 such234

that we can make fair comparison of pure spatial effects in the later sections.235

We do not impute locations with r < 100 m. We do not have any data in that236

region and any inference relies on extrapolation.237

2.5.2 Excess Pb inside the plume238

The plume is a cone defined by C = {θ : 260° < θ < 310°}. At each distance r,we compute the plume excess, the difference of soil Pb (ppm) between the plumes on theoutside along the radius r ring, by

Excessf (r) =

∑j:θj∈C f(θj , r)∑

j:θj∈C 1−∑j:θj 6∈C f(θj , r)∑

j:θj 6∈C 1, (3)

–8–


where f(θj , r) is a posterior predictive draw of f at location θj , r using (1). Because it239

is calculated using posterior draws, expression (3) is a random variable whose posterior240

distribution we can summarize using the mean or quantile of its simulation draws.241

Likewise we compute the excess Pb at the observational level,

Excessy(r) =

∑j:θj∈C y(θj , r)∑

j:θj∈C 1−∑j:θj 6∈C y(θj , r)∑

j:θj 6∈C 1. (4)

where y(θj , r) denotes a posterior predictive draw of y|f(θj , r) using (2). In general, Excessy(r)242

has a lager posterior variance. It is because of the noise ε: such that there are more un-243

certainty even if we repeat sampling in the same location. It also has larger posterior244

mean than Excessf (r). This is due to the multiplicative measurement error in ε. For ex-245

ample, if there is a multiplicative noise source that will halve or double the true value246

f with equal probability, the posterior mean of y becomes 2+0.52 f = 1.25f .247

We further aggregate the the excess amount of Pb in the plume within the circle248

of radius r. To this end, we reweigh the excess density of the ring by its radius,249

Excessfcircle(r) =

∫ r

0

r′Excessf (r′)dr′, (5)

and

Excessycircle(r) =

∫ r

0

r′Excessy(r′)dr′. (6)

Finally we estimate the excess amount of Pb of the plume inside a circle with anygiven radius r by

|C|360× πr2 × thickness× soil density × Excessycircle(r). (7)

For all the above quantities we compute the posterior mean, 50%, and 95% inter-250

vals using the simulation draws.251

3 Results252

3.1 Raw data summary253

Concentrations of Pb measured in all surface soil samples range from 30 to 9,000254

mg/kg and average 400 mg/kg (median of 140 mg/kg). All four Pb concentrations >2000255

mg/kg are inside the plume area and within a distance of 400 m from the cathedral. Over-256

all, soil Pb concentrations average 200 mg/kg outside (n = 45) and 400 mg/kg (n =257

55) inside the plume area, respectively (Fig. 2). Average soil Pb for tree pits (300 mg/kg;258

n = 55) and garden areas (500 mg/kg; n = 7) are comparable, but markedly lower259

in park areas (130 mg/kg; n = 23). Cracks in the sidewalk (1400 mg/kg, n = 13) on260

the other hand are often higher in Pb than neighboring tree pits and garden areas (Fig.261

3). Among the other soil constituents analyzed by XRF, only Sn shows a systematic re-262

lationship with Pb at higher concentrations. For 8 samples in the 1000-9000 mg/kg range263

of Pb concentrations, the mass ratio of Sn relative to Pb averages 3.5% after recalibrat-264

ing the XRF signal.265

Unlike air, water, and food, there is no standard in France for the Pb content of266

soil in outdoor public areas. A recommendation from French health authorities of 300267

mg/kg corresponds approximately to the level at which blood-Pb of 5% of infants com-268

ing in contact with the soil would exceed a threshold of 50 µg/L (HCSP, 2014). For com-269

parison, the current US Environmental Protection Agency standard for residential soil270

–9–


43

73

104

134

134

165

195

226

256

287

317

348

378

409

439 ND

Posterior mean of soil Pb (ppm)

W, 1.5 km E, 1.5 km

S, 1.5 km

N, 1.5 km

500 m 1 km

500 m

1 km102

103

104

observed samples

(ppm)

140 180

180

220

220

260

260

300

300

340 340

380

380

420

420

90% quantile of soil Pb

90

126 161

161

196

231

267

267

302

337

478


30

30

55 81

81

106

132

157

183

208

234

311


23

23

45 67

89

112

134

156

178

201 245


Figure 5: Contour plot of posterior mean and quantiles of f (net of soil types and mea-surement errors) of soil Pb concentrations within 1.5 km of the cathedral.

in areas where children play is 400 mg/kg Pb, but lowering this value is under discus-271

sion. Relative to 300 mg/kg, the Pb content of 29 of our 100 samples exceeds the French272

recommended value, 21 of which inside the plume area and 8 outside (Fig. 3). Consid-273

ering only the samples collected along the two concentric circles to avoid bias, the av-274

erage soil Pb content within the plume is 500±200 mg/kg (n = 7, 1 sd), compared to275

200±40 mg/kg (n=51) outside the plume (Fig. 4).276

3.2 Model inference277

Contours of modeled Pb concentrations show more elevated levels in a northwest-278

erly direction from the cathedral compared to other areas (Fig. 5). Although this peak279

was identified independently, it corresponds closely to the direction of the plume derived280

from meteorological observations (INERIS, 2019). Bayesian inference encodes all uncer-281

tainty, which is displayed as 90%, 75%, 25%, 10% quantiles of the predicted spatial con-282

centration of soil Pb concentrations f(r, θ), at all imputed locations among the 1.5 km283

neighborhood around the cathedral (Fig. 5). The estimation separates all measurement284

errors and soil types. In locations where not enough data is collected nearly (south, east),285

the model essentially has to extrapolate, but the posterior standard deviation of f is also286

large.287

The effect of soil types indicates a decline in Pb concentrations from cracks in the288

sidewalk to tree pits and parks (Fig. 6 ). The effect of areas described as gardens is more289

variable, and poorly constrained in the two cases of the two plant pots. The number is290

on the scale of y1/4, and for the median values y ≈ 140, an additive effect of 0.5(−0.5)291

on y1/4, which corresponds to 100(−65) ppm increase on y.292

The model estimates Pb concentrations f as a function of the bearing from the cathe-293

dral, evaluated at distances of 400 m and 1000 m and average concentrations over all dis-294

tances <1.5 km (Fig. 7). At the 400 m ring, the soil Pb for outside-plume-average is about295

190 (95% interval, 130-270) mg/kg, and peaks at 490 (95% interval, 330-710) mg/kg in296

the core of the plume.297

The posterior predictive distribution of Excessf (r) (Eqn. 3) shows that the differ-298

ence in Pb concentration between inside and outside the plume declines from 350 (95%299

–10–


soil type effect

−1.5 −1.0 −0.5 0.0 0.5 1.0 1.5

park

plant pot

tree

garden

crack

Figure 6: Posteriormean, and 95% in-tervals for soil typeeffects µk.

Comparison across angles by averaging over all distances

0

200

400

600

0 N

90 E

180 S

270 W

360 N

angle

soil Pb ppm

plume

Comparison across angles at distance near 400 m (300−500)

0

200

400

600

0 N

90 E

180 S

270 W

360 N

angle

plume

Comparison across angles at distance near 1000 m (900−1100)

0

200

400

600

0 N

90 E

180 S

270 W

360 N

angle

plume

Figure 7: Modeled Pb concentrations as a function of directionin relation to the cathedral, evaluated at distances of 400 m and1000 m and averaged over all distances <1.5km.

0

500

1000

1500

distance (m)

0 500 1000 1500

Posterior draws of f inside plumes

soil Pb

0

500

1000

1500

distance (m)

0 500 1000 1500

Posterior draws of f outside plumes

Comparison of f between inside and outside plumes

−100

0

200

400

600

0 500 1000 1500

distance (m)

soil Pb inside minus

outside

Comparison of y between inside and outside plumes

−300

0

300

600

900

0 500 1000 1500

distance (m)

soil Pb inside

dividing outside

Figure 8: From left, 1-2: Posterior draws of mean Pb concentrations f inside and out-side plumes as a function of distance from the cathedral. The uncertainty is large smalland large distance where there is no nearby data. 3: The posterior mean, of the differencebetween and inside and outside the plume (Eqn. 3). 4: The comparison of predicted Pbconcentrations to be observed y (Eqn. 4). With additional observational noise added, theuncertainty interval is much wider.

interval, 140-640) mg/kg at 200 m from the cathedral to 200 (95% interval, 90-330) mg/kg300

at 500 m, and 90 (95% interval, 0-190) mg/kg at 900 m, respectively, and vanishes be-301

yond that distance (Fig. 8).302

The model also calculates the average excess Pb inside a given radius (Eqn. 5–6)303

on both the mean response f and with additional observational noise respectively (Fig.304

9). On the observational level y, inside the 1 km circle, the average concentration of Pb305

inside the plume is 430 (95% interval, 300-590) mg/kg is nearly double the average Pb306

concentration in the other directions of 240 (95% interval, 170-320) mg/kg. Finally, the307

model calculates the corresponding integrated mass of excess is 1.0 (95% interval, 0.5-308

1.5) metric tons of Pb at a 1000 m distance from the cathedral and becomes poorly con-309

strained beyond that distance for lack of data (Fig. 9).310

4 Discussion311

Soil Pb concentrations around Notre-Dame cathedral show considerable spatial vari-312

ability, both inside and outside the plume area. In some cases, this may reflect site-specific313

factors such as newly added soil (Fig. 6). This may be why park areas are generally lower314

in Pb. Cracks in the sidewalk, on the other hand, are generally higher in Pb possibly be-315

cause of a preserved legacy of contamination from decades of leaded gasoline use. An oc-316

–11–


weighted excess of f (ppm)

−100

0

200

400

600

0 500 1000 1500radius (m)

94

soil Pb ppm

weighted excess of y (ppm)

−200

0

300

600

900

0 500 1000 1500radius (m)

116

excess of pb (in f) inside the circle (kg)

0

1000

2000

3000

4000

0 500 1000 1500

1335

radius (m)

excess of pb (in y) inside the circle (kg)

0

1000

2000

3000

4000

0 500 1000 1500

1645

radius (m)

Figure 9: From left, 1-2: The average excess Pb (Eqn. 5–6) inside the radius-r-circle, ofthe mean response f and observation y. 3-4: The accumulated exceed Pb (Eqn. 7) insidethe radius-r-circle.

casional highly local source of contamination from Pb paint or other sources cannot be317

ruled out, although these were apparently not sufficient to erase a pattern that is con-318

sistent with the trajectory of the plume. The model effectively subtracts systematic dif-319

ferences in background Pb concentrations for different soil types when calculating the320

excess inside the plume to outside the plume.321

The Pb tiles covering the roof of the cathedral and the spire date to the second half322

of the 19th century (Daly, 1866). Some combination of Sn and Pb in solder was prob-323

ably used extensively to cover the roof and spire of the cathedral. The relatively con-324

stant proportion of Sn relatively to Pb in the soil with very high levels of Pb can there-325

fore be attributed to the fire. Concentrations of Sn relative to Pb are not sufficiently el-326

evated, however, to separate different sources of Pb at lower levels of contamination.327

The background level below 200 mg/kg Pb outside the plume is plausible given av-328

erage crustal Pb concentrations of <100 mg/kg with perhaps some legacy of leaded-gasoline329

use until 2000 (Miquel, 2001). Without the model, the difference in Pb concentrations330

between the area inside and outside the plume would have been poorly constrained (Fig.331

8). A key question is the extent to which this excess is representative of the overall fall-332

out over the plume area, including hard surfaces such as sidewalks and roads where this333

excess could have been washed away by rain. Only 3 mm of rain was recorded during334

the week following the fire, but a total of 92 mm fell over Paris within 4 four weeks of335

the fire (https://www.historique-meteo.net/france/ile-de-france/paris/2019/336

04/).337

Lead has a particularly strong tendency to adsorb to mineral surfaces (Selim, 2017).338

Once in contact with soil, Pb is therefore unlikely to be flushed off particles by water,339

especially within less than a year, unless by physical removal of the soil. If anything, tree340

pits might be concentrating Pb from a larger area if surface runoff percolates through341

tree pits and supplies particles from nearby hard surfaces. However, this also seems un-342

likely given the extensive drainage system along the sides of the streets of Paris, which343

is lower in elevation than the sampling sites.344

A more likely mechanism for concentrating Pb in tree pits is capture of airborne345

particles by tree leaves, followed by rainfall rinsing the leaves or settling of the leaves into346

the tree pit. Studies of the natural radioisotope 210Pb, whose atmospheric fallout is known,347

have shown that this process can enhance its accumulation by one- to two-thirds under348

the canopy of trees (Fowler et al., 2004), but not by an order of magnitude. Parks with-349

out trees, on the other hand, should not be subject to this process and might be more350

indicative of the fallout, at least in the short term and before erosion or the addition of351

new soil.352

–12–


For comparison of our estimate of 1000 kg of excess Pb deposited downwind of the353

fire, the 50 km-long plume emanating from fire beyond a distance of 1 km was estimated354

to contain about 150 kg Pb on the basis of a furnace experiment using a combination355

metallic Pb and plastic (INERIS, 2019). Whereas the possibility of preferential accumu-356

lation of Pb in tree pits cannot be ruled out, the amount of Pb deposited within 1000357

m of the cathedral estimated from the soil survey is fairly well constrained. About 6 times358

more Pb was therefore deposited within 100-1000 m of the cathedral than beyond that359

distance. For perspective, the addition of Pb to gasoline resulted in air emission of 4100360

tons of Pb per year in France in 1990 (Miquel, 2001). Using population as a proxy for361

traffic and accounting for the one-fifth proportion of the French population residing in362

the greater Paris region, this suggests that the population of the city was exposed at the363

time to emissions of about 800 tons of Pb every year. Leaded gasoline was banned in 2000364

and airborne emissions of Pb have dropped by at least an order of magnitude since Motelay-365

Massei et al. (2005). The impact of the Notre-Dame fire would therefore have been dwarfed366

by the impact of automobile traffic a few decades ago, and much harder to detect in soil.367

A puzzle arises when the average excess of 200 ppm/kg Pb in the plume is converted,368

using our approximate sampling depth of 1 cm, to 4,000,000 µg/m2 Pb, the unit and type369

of measurement more frequently referred to in regulation of indoor surfaces, including370

in schools. Such very high levels are reported on the interactive ARS map only within371

100 m of the cathedral itself, in an area that was still out of bounds for the general pub-372

lic as of May 2020. At greater distances, but still within 1000 m of the cathedral, reported373

values are all below 20,000 µg/m2. Many of the reported measurements, however, date374

from summer 2019 or later by which time much of the Pb fallout could have been flushed375

off hard surfaces such sidewalks and roadways by rain or washing. Even if our soil Pb376

measurements could overestimate the overall Pb fallout by a factor of 2 because of lo-377

cal concentration, it appears likely that the measurements based on outdoor surface wipes378

reported by the government considerably underestimate the amount of the Pb that was379

actually deposited in the plume area because of their timing. Concentrations of Pb on380

hard surfaces are likely to return more rapidly to background than in soil, whose retained381

inventory is therefore likely to provide a more accurate record of the fallout from the fire.382

What are the implications of the soil-based findings for human exposure in the plume383

area in the aftermath of the fire, especially for small children who are most vulnerable?384

No children are likely to play around the tree pits themselves or even the sampling sites385

designated as gardens, many of which are not suitable playing areas (see interactive map386

with photos listed under the Acknowledgments). Fortunately, the more likely playing ar-387

eas such as parks were generally low in Pb (Figs. 4, 6). The potential source of expo-388

sure therefore lies elsewhere and would have been the dust deposited during and imme-389

diately after the fire. This impact is difficult to ascertain independently for lack of spe-390

cific information about in-house swipe measurements and a sufficient number of timely391

blood Pb measurements (Fig. 1). Unlike in New York City for instance, infants are not392

systematically tested for blood Pb in France and their exposure before the fire is there-393

fore also not well known.394

Seven weeks after the cathedral fire, local authorities offered to test children from395

volunteer families, but the number of tests remained very low through July, 2019. Af-396

ter exposure ends, blood-Pb levels can decline within a few weeks although it can also397

take much longer (Barbosa et al., 2005). The low proportion (1%) of children reported398

with blood-Pb levels >50 µg/L is welcome news but may mask a temporarily much higher399

level of exposure in the days to a few weeks after the fire. The few cases of surfaces el-400

evated in Pb reported for schools in the affected area also date from summer 2019 and401

therefore likely underestimate peak exposure in the 1-2 weeks following the fire, espe-402

cially if the schools had followed earlier recommendations and already cleaned the com-403

mon areas. Finally, because the blood survey was relying on volunteers instead of pro-404

actively seeking all 6,000 potentially exposed children in the affected area through a door-405

–13–


to-door survey, it was probably biased towards a more educated, wealthier segment of406

the population that may have been less at risk. In a post-coronavirus world, the need407

and feasibility of a testing campaign of the magnitude commensurate with the scale of408

a large fire or other environmental accident has become much harder to argue against.409

5 Conclusions410

A report issued by the ARS (2019h) on April 16, 2020, exactly one year after the411

fire, acknowledges the possibility that more people than indicated by the available data412

may have been exposed to Pb as a result of the cathedral fire. Our observations support413

this scenario. The soil data show that the levels of Pb contamination expressed in terms414

of mass per unit area documented within 100 m of the cathedral during summer 2019,415

several months after the fire, are comparable to the Pb fallout that extended downwind416

of the cathedral to a distance of 1 km. Therefore, elevated levels of Pb in indoor dust417

probably extended up to 1 km from the cathedral as well.418

From a policy perspective, the tracking of the impact of Pb from the Notre-Dame419

fire suggests that the administration of large cities such as Paris should have a large en-420

vironmental investigation team on standby, ready to be deployed to make hundreds of421

measurements immediately after an accident or toxic spill that could potentially pose422

a threat to public health. The city of Paris in fact has such a team (http://laboratoirecentral423

.interieur.gouv.fr/Presentation/Le-LCPP/Panorama), which was deployed and col-424

lected Pb data after the fire, but evidently not rapidly enough or at the required scale.425

The results from such an environmental investigation should be communicated imme-426

diately in ways that allow the general public to know exactly where the hazards are, which427

is very easy today using the mapping function of smartphones. In addition, public health428

authorities should be more prepared to survey immediately with environmental testing429

and biomarkers measurements by going door to door instead of waiting for volunteers430

and, again, rapidly communicate those results while providing the necessary privacy pro-431

tection.432

Acknowledgments433

This project was supported in part by NIEHS P42 grant ES010349 and NSF grant CNS-434

1730414. A. Casella and C. van Geen participated in the soil sampling. B. Bostick, S.435

Chillrud, and B. Mailloux provided helpful suggestions for interpreting the soil data. Ex-436

changes with A. Lefranc, R. Charvet, P. Glorennec, and P. Garnoussi in France pointed437

us to publicly available data and gave us a perspective of the activities conducted by French438

authorities in the aftermath of the fire. The entire data set and an interactive map of439

the test results with photos of each sampling site are available at https://shorturl.at/440

kuvD5, and the replication R and Stan code at https://github.com/yao-yl/parisPb.441

The data will also be made available through an approved repository upon acceptance.442

Reference443

Agence Regionale de la Sante (2019a). https://www.pressesante.com/wp-content/444

uploads/2019/04/CP-PP-27042019-Information-aux-riverains-de-Notre-Dame445

-1-3.pdf446

Agence Regionale de la Sante (2019b). https://www.iledefrance.ars.sante.fr/incendie447

-notre-dame-information-aux-riverains-sur-les-consequences-des-retombees448

-de-plomb449

Agence Regionale de la Sante (2019c). https://www.iledefrance.ars.sante.fr/suivi450

-des-recommandations-sanitaires-suite-lincendie-de-la-cathedrale-notre451

-dame-de-paris452

–14–


Agence Regionale de la Sante (2019d). https://www.iledefrance.ars.sante.fr/system/453

files/2019-07/DP-point-de-situation-Notre-Dame-de-Paris 0.pdf454

Agence Regionale de la Sante (2019e). https://www.iledefrance.ars.sante.fr/mise455

-au-point-de-lars-ile-de-france-suite-un-article-de-mediapart456

Agence Regionale de la Sante (2019f). https://www.iledefrance.ars.sante.fr/suivi457

-des-consequences-de-lincendie-de-notre-dame-point-de-situation-au-27112019458

Agence Regionale de la Sante (2019g). https://www.iledefrance.ars.sante.fr/system/459

files/2019-10/PE Plombemies ND 10%2010%202019.pdf460

Agence Regionale de la Sante (2019h). https://www.iledefrance.ars.sante.fr/incendie461

-de-notre-dame-de-paris-bilan-sanitaire-un462

Aizer, A. & Currie, J. (2019). Lead and Juvenile Delinquency: New Evidence from Linked463

Birth, School and Juvenile Detention Records. The Review of Economics and Statis-464

tics, 101, 575-587.465

Alexievich, S. (2006) Voices from Chernobyl: The Oral History of a Nuclear Disaster,466

Picador, USA.467

ANSES (2020). La contamination d’espaces publics exterieurs par le plomb. Agence na-468

tionale de securite sanitaire de l’alimentation, de l’environnement et du travail. Sai-469

sine n° 2019-SA-0147. https://www.anses.fr/fr/system/files/AIR2019SA0147470

.pdf471

Barbosa Jr, F., Tanus-Santos, J. E., Gerlach, R. F., & Parsons, P. J. (2005). A critical472

review of biomarkers used for monitoring human exposure to lead: Advantages, lim-473

itations, and future needs. Environmental Health Perspectives, 113, 1669-1674.474

Brown, A., Franken, P., Bonner, S., Dolezal, N., & Moross, J. (2016). Safecast: success-475

ful citizen-science for radiation measurement and communication after Fukushima.476

Journal of Radiological Protection, 36,, S82–S101.477

Cheng, Z., Zheng, Y., Mortlock, R., & van Geen, A. (2004). Rapid multi-element anal-478

ysis of groundwater by high-resolution inductively coupled plasma mass spectrom-479

etry. Analytical and Bioanalytical Chemistry, 379, 513-518.480

Chillrud, S. N., Bopp, R. F., Simpson, H. J., Ross, J. M., Shuster, E. L., Chaky, D. A.,481

Walsh, D. C., Choy, C. C., Tolley, L. -R., & Yarme, A. (1999). Twentieth century482

atmospheric metal fluxes into Central Park Lake, New York. Environmental Sci-483

ence & Technology, 33, 657-662.484

Daly, C. (1866). La couverture de Notre-Dame de Paris. Revue Generale de l’Architecture485

et des Travaux Publics, 14, 211-212.486

Etchevers, A., Le Tertre, A., Lucas, J-P., Bretin, P., Oulhote, Y., Le Bot, B., & Gloren-487

nec, P. (2015). Environmental determinants of different blood lead levels in chil-488

dren: A quantile analysis from a nationwide survey. Environment International,489

74, 152-159.490

Fowler, D., Skiba, U., Nemitz, E., Choubedar, F., Branford, D., Donovan R., & Row-491

land, P. (2004). Measuring Aerosol and Heavy Metal Deposition on Urban Wood-492

land and Grass Using Inventories of 210Pb and Metal Concentrations in Soil. Wa-493

ter, Air, & Soil Pollution: Focus 4, 483–499.494

–15–


Glorennec, P., Lucas, J.P., Mercat, A.C., Roudot, A.C., & Le Bot, B. (2016). Environ-495

mental and dietary exposure of young children to inorganic trace elements. Envi-496

ronment international, 97, 28-36.497

HCSP (2014). Expositions au plomb: determination de nouveaux objectics de gestion.498

Haut Conseil de la Sante Publique. www.hcsp.fr499

JORF (2009). Arrete du 12 mai 2009 relatif au controle des travaux en presence de plomb,500

realises en application de l’article L. 1334-2 du code de la sante publique. Journal501

Officiel de la Republique Francaise https://www.legifrance.gouv.fr/affichTexte502

.do?cidTexte=JORFTEXT000020668963&categorieLien=id503

Laidlaw M. A., & G. M. Filippelli, G. M. (2008). Resuspension of urban soils as a per-504

sistent source of lead poisoning in children: A review and new directions. Applied505

Geochemistry, 23, 2021-2039.506

Landes, F., Paltseva, A., Sobolewski, J., Cheng, Z., Ellis, T., Mailloux, B., & van Geen,507

A. (2019). A field procedure to screen soil for hazardous lead. Analytical Chem-508

istry, 91, 8192-8198.509

Lanphear B. P., Emond M., Jacobs D.E., Weitzman M., Tanner M., Winter N. L., Yakir510

B., & Eberly S. (1995). A side-by-side comparison of dust collection methods for511

sampling lead-contaminated house dust. Environmental Research, 68, 114–23.512

Lanphear, B. P., Hornung, R., Khoury, J., Yolton, K., Baghurst, P., Bellinger, D. C., Can-513

field, R.L., et al. (2005). Low-level environmental lead exposure and children’s in-514

tellectual function: an international pooled analysis. Environmental Health Per-515

spectives, 113, 894-899.516

Le Monde (2019). Apres l’incendie de Notre-Dame, une plainte contre la pollution au517

plomb. https://www.lemonde.fr/planete/article/2019/07/29/pollution-au518

-plomb-autour-de-notre-dame-plainte-contre-x-pour-mise-en-danger-d-autrui519

5494472 3244.html520

Mediapart (2019). Notre-Dame: des taux de plomb chez les enfants superieurs au seuil521

de vigilance. https://www.mediapart.fr/journal/france/050819/notre-dame522

-des-taux-de-plomb-chez-les-enfants-superieurs-au-seuil-de-vigilance523

?onglet=full524

Miquel, G. (2001). Rapport sur les effets des metaux lourds sur l’environnement et la525

sante. Office parlementaire d’evaluation des choix scientifiques et technologiques.526

French Senate report. https://www.senat.fr/rap/l00-261/l00-26150.html527

Motelay-Massei, A., Ollivon, D., Tiphagne, K., & B. Garban, B. (2005). Atmospheric528

bulk deposition of trace metals to the Seine river Basin, France: concentrations,529

sources and evolution from 1988 to 2001 in Paris. Water Air Soil Pollut, 164, 119–135.530

Robin des Bois (2019). Ile de la Cite: a new polluted site in Paris. https://robindesbois531

.org/notre-dame-un-nouveau-site-pollue-a-paris/532

Selim, H. M. (2017). Competitive Sorption and Transport of Heavy Metals in Soils and533

Geological Media. CRC Press, Boca Raton, Florida, USA.534

Stan Development Team (2018). Stan Modeling Language Users Guide and Reference535

Manual, Version 2.18.0. http://mc-stan.org536

–16–

fallout of lead over paris from the 2019 notre-dame...

Documents